Table of Contents



Partners

ELDA

ELDA (Evaluations and Language Resources Distribution Agency) was created in February 1995, with the aim to identify, classify, collect, validate and produce the language resources which may be needed by the HLT – Human Language Technology – community.

Along the years, ELDA enlarged its activities to language technology evaluation and to the distribution of Language Resources needed for evaluation works. Today, the firm is structured into two departments: Language Resource Identification and Distribution and Language Resource Production & Evaluation.

ELDA activities are supervised by its general director and founder of the firm, Khalid Choukri. To this date, about twelve persons are employed on a full time basis. From time to time, other people can come to add some efforts on temporary actions or specific projects.

Within the past years, ELDA has been involved in many projects and evaluation campaigns, not only at the national level but also at the European level. In particular, ELDA participated to international projects, such as TC-STAR and CHIL. The agency has also been involved in the CLEF project (Cross Language Evaluation Forum). At the national leval, ELDA took part to a great number of projects within the big French national programme Technolangue, which included 8 evaluation campaigns.

Contact: Djamel Mostefa

LIG/GETALP

The GETALP (Groupe d'Etude en Traduction/Traitement des Langues et de la Parole) from the new LIG laboratory (Laboratoire d’Informatique de Grenoble) results from the fusion of GEOD and GETA teams of the former CLIPS laboratory. Consequently, the research topics of this new team are very large from speech analysis to natural language processing with core activities in automatic speech recognition and machine translation in a multilingual framework.

Contact: Laurent Besacier

LIA

LIA is the computer laboratory of the University of Avignon (UAPV). LIA carries on research on Natural Language Processing, Speech Processing, Communications, and Optimization.

In the language processing field the main topics are speech recognition, speaker recognition, spoken dialog, automatic summarization and multimedia information retrieval.

LIA researchers serve on the Editorial Board of international journals, including Computational Intelligence, Computer Speech and Language, Speech Communications, Pattern Recognition Letters. They have also served on the scientific committees of many international conferences. LIA has been active in various corpus based evaluation programs including ESTER (French evaluation program for broadcast news speech transcription, segmentation, speaker tracking and named entities extraction), MEDIA (understanding in spoken dialogue), EQUER (question answering), NIST (speaker recognition, Rich Transcription) and TREC (information retrieval). LIA is or has been involved in several projects funded at the European or French level like MTM (PCRD, speech and speaker recognition), COST 275, BioSecure (NOE), SMADA (Speech Driven Multi-Modal Automatic Directory Assistance) IST project, PASCAL (European Network of Excellence) and the IST European project DIVINES. LIA is also involved on several research contracts with Industries including FT and Thales.

Contact: Fabrice Lefevre

LIUM

LIUM is the Computer Science Laboratory of the Université du Maine (Le Mans, France). The LIUM's competence concerned by the PORT-MEDIA project is about the automatic speech processing. Participating to the French ESTER evaluation campaign, LIUM has developed very competitive entire automatic transcription and audio/speaker segmentation systems [Deleglise 05, ESTER 05]: they reached the second position on TRS (transcription) and SRL (speaker diarization) ESTER tasks. The LIUM ASR system was the best open source ASR system during this evaluation. Major part of these systems are available on the LIUM website under open source license (http://www-lium.univ-lemans.fr/speechtools).

The LIUM transcription system is based on the CMU Sphinx s3.3 decoder: LIUM has provided to the CMU Speech Group (Carnegie Mellon University) some tools which are yet integrated on the last canonical version (s3.6) of the Sphinx 3 decoder, and other tools provided by LIUM should be integrated on future versions of Sphinx 3 decoder. The LIUM transcription system represents a very good base to work on the PORT-MEDIA project.

Contact: Yannick Esteve

LORIA

The LORIA/TALARIS Team has been developing high expertise in Natural Language Processing for years now, on various levels : parsing, generation, dialogue studies, dialogue systems and linguistic resources elaboration and normalization (lexicon, grammars, semantic). During the last 4 years, a dialogue system has been developed and evaluated in the MEDIA project, giving very interesting results despite the difficult challenge of understanding agrammatical utterances with deep-parsing techniques, usually designed for textual analysis. This challenge has been partially overcome by the help of logics at semantic level (based on ontologies). In the current proposal, TALARIS will work on transcoding semantic information given by statistical system to a higher level, more complex and more complete, using logical inference and ontologies, just like in the previous project, but in the reverse direction.

Contact: Matthieu Quignard

Relevance of the partners

The partners involved in PORT-MEDIA have shown high experience in all these aspects:

All partners have their system main components based on open-source tools (LIUM's ASR system is based on the SPHINX toolkit, LIA's SLU system is based on GMTK and CRF++, LIG's MT system is based on Moses/Giza++ tools,…) which will ensure the reproducibility of the proposed methods

Finally, LIA, LORIA and ELDA have all previously participated at the Technolangue MEDIA project; therefore they have played a central role in the definitions of the protocol used during the evaluation.

More on ELDA: ELDA has for objective to identify, classify, collect, validate and produce the language resources which may be needed by the HLT – Human Language Technology – community. Besides, ELDA is involved in HLT evaluation campaigns. Thus, ELDA launched several evaluation activities for some years, by distributing language resources convenient to evaluation experiments in human language technology. To validate this new activity, ELDA changes the meaning of its name « European Language Resources Distribution » into « Evaluation and Language Resources Distribution Agency ». ELDA participated to many European evaluation projects, such as CLEF and Amaryllis, and to many national projects within the Technolangue programme. ELDA’s involvement to many evaluation campaigns allows to guarantee a joint action around resources between several distinct campaigns. Moreover, in the same way as ELDA made a great number of language resources available and carries on to its distribution activities in an increasing way, it also aims to act as one of the European expert centre in several domains, as its mission was with respect to Language Resources in the ‘90s. Knowledge capitalization within ELDA, in terms of campaign organisation, must enable first to re-initiaite any evaluation campaign upon demand. Then, ELDA wishes to be able to distribute the produced resources or resources enriched within the campaign, to guaranty their continuous availability, as well as a “all in one” evaluation distribution, which allows all players in the field to reproduce an evaluation at any moment on a required topic.

Previous collaborations

All partners have already collaborated in (at least) one project with one other partner:

Fabrice Lefevre 2010/02/02 13:57 custom essay writing services Knee Pain blog free dating site write my essay winrar скачать business commercial loans